Go package for loading program configuration from multiple sources.
Loading program configurations is usually done by parsing the arguments passed
to the command line, and in this case the standard library offers a good support
with the flag
package.
However, there are times where the standard is just too limiting, for example
when the program needs to load configuration from other sources (like a file, or
the environment variables).
The conf
package was built to address these issues, here were the goals:
-
Loading the configuration has to be type-safe, there were other packages available that were covering the same use-cases but they often required doing type assertions on the configuration values which is always an opportunity to get the program to panic.
-
Keeping the API minimal, while the
flag
package offered the type safety we needed it is also very verbose to setup. Withconf
, only a single function call is needed to setup and load the entire program configuration. -
Supporting richer syntaxes, because program configurations are often generated dynamically, the
conf
package accepts YAML values as input to all configuration values. It also has support for sub-commands on the command line, which is a common approach used by CLI tools. -
Supporting multiple sources, because passing values through the command line is not always the best approach, programs may need to receive their configuration from files, environment variables, secret stores, or other network locations.
A program using the conf
package needs to declare a struct which is passed to
conf.Load
to populate the fields with the configuration that was made
available at runtime through a configuration file, environment variables or the
program arguments.
Each field of the structure may declare a conf
tag which sets the name of the
property, and a help
tag to provide a help message for the configuration.
The conf
package will automatically understand the structure of the program
configuration based on the struct it receives, as well as generating the program
usage and help messages if the -h
or -help
options are passed (or an error
is detected).
The conf.Load
function adds support for a -config-file
option on the program
arguments which accepts the path to a file that the configuration may be loaded
from as well.
Here's an example of how a program would typically use the package:
package main
import (
"fmt"
"github.com/segmentio/conf"
)
func main() {
var config struct {
Message string `conf:"m" help:"A message to print."`
}
// Load the configuration, either from a config file, the environment or the program arguments.
conf.Load(&config)
fmt.Println(config.Message)
}
$ go run ./example.go -m 'Hello World!'
Hello World!
By default, conf
will look for environment variables before loading command-line configuration flags with one important caveat: environment variables are prefixed with the program name. For example, given a program named "foobar":
func main() {
config := struct {
Name string `conf:"name"`
}{
Name: "default",
}
conf.Load(&config)
fmt.Println("Hello", config.Name)
}
The following will be output:
$ ./foobar // "Hello default"
$ FOOBAR_NAME=world ./foobar // "Hello world"
$ FOOBAR_NAME=world ./foobar --name neighbor // "Hello neighbor"
$ MAIN_NAME=world go run main.go // "Hello world"
If you want to hard-code the prefix to guarantee immutability or just to customize it, you can supply a custom loader config:
loader := conf.Loader{
Name: "my-service",
Args: os.Args[1:],
Sources: []conf.Source{
conf.NewEnvSource("MY_SVC", os.Environ()...),
},
}
conf.LoadWith(&config, loader)
While the conf.Load
function is good enough for common use cases, programs
sometimes need to customize the default behavior.
A program may then use the conf.LoadWith
function, which accepts a
conf.Loader
as second argument to gain more control over how the configuration
is loaded.
Here's the conf.Loader
definition:
package conf
type Loader struct {
Name string // program name
Usage string // program usage
Args []string // list of arguments
Commands []Command // list of commands
Sources []Source // list of sources to load configuration from.
}
The conf.Load
function is actually just a wrapper around conf.LoadWith
that
passes a default loader. The default loader gets the program name from the first
program argument, supports no sub-commands, and has two custom sources setup to
potentially load its configuration from a configuration file or the environment
variables.
Here's an example showing how to configure a CLI tool that supports a couple of sub-commands:
package main
import (
"fmt"
"github.com/segmentio/conf"
)
func main() {
// If nil is passed instead of a configuration struct no arguments are
// parsed, only the command is extracted.
cmd, args := conf.LoadWith(nil, conf.Loader{
Name: "example",
Args: os.Args[1:],
Commands: []conf.Command{
{"print", "Print the message passed to -m"},
{"version", "Show the program version"},
},
})
switch cmd {
case "print":
var config struct{
Message string `conf:"m" help:"A message to print."`
}
conf.LoadWith(&config, conf.Loader{
Name: "example print",
Args: args,
})
fmt.Println(config.Message)
case "version":
fmt.Println("1.2.3")
}
}
$ go run ./example.go version
1.2.3
$ go run ./example.go print -m 'Hello World!'
Hello World!
We mentioned the conf.Loader
type supported setting custom sources that the
program configuration can be loaded from. Here's the the conf.Source
interface
definition:
package conf
type Source interface {
Load(dst Map)
}
The source has a single method which receives a conf.Map
value which is an
intermediate representation of the configuration struct that was received by the
loader.
The package uses this type internally as well for loading configuration values
from the program arguments, it can be seen as a reflective representation of the
original value which exposes an API that is more convenient to use that having
a raw reflect.Value
.
One of the advantages of the conf.Map
type is that it implements the
objconv.ValueDecoder
interface and therefore can be used directly to load configurations from a
serialized format (like JSON for example).
Last but not least, the conf
package also supports automatic validation of the
fields in the configuration struct. This happens after the values were loaded
and is based on gopkg.in/validator.v2.
This step could have been done outside the package however it is both convenient and useful to have all configuration errors treated the same way (getting the usage and help message shown when something is wrong).